What models are included in the Hugging Face Collection on Foundry?

The collection includes trending models from the Hugging Face ecosystem across every modality — text, vision, audio, and multimodal — such as LLMs and VLMs for chat and agents, ASR and speech translation, embeddings, segmentation, and image generation. The catalog is refreshed weekly as the community publishes new models.

How does Microsoft ensure security for these models?

Every model in the collection is security-screened, ships in SafeTensors weight format with no trust_remote_code execution paths unless rigorously reviewed, and Microsoft builds inference container images on supported runtimes that are scanned for CVEs, signed, and published to a Microsoft-managed registry.

What happens to model weights, and where are they stored?

Model weights are pulled from Hugging Face once, validated against the published model card, and stored in Microsoft-managed Azure storage in the regions where the model is served.

Open-Source AIHugging Face BlogJul 7, 2026Jul 8, 2026, 01:00 JST

Microsoft Brings Hugging Face Open Models to Foundry Platform

Send on LINE

3 Key Points

What happened
Microsoft announced at Build 2026 that Hugging Face open-weight models are now available on Foundry Managed Compute — a curated catalog refreshed weekly and deployable in one click, with weights pre-staged in Azure and runtimes built and scanned by Microsoft.
Why it matters
Open models have closed the gap with proprietary models on benchmarks, and this integration removes the operational burden — license review, security screening, runtime selection, GPU sizing, and CVE patching — that has historically made deploying them enterprise-grade difficult. Developers can now run open models on infrastructure they control, with the same endpoint, SDKs, authentication, and observability as frontier models.
What to watch
The catalog spans every modality — text, vision, audio, and multimodal — with runtimes including vLLM, SGLang, TensorRT-LLM, NIM, TEI, and llama.cpp. Every model ships in SafeTensors format with no untrusted code execution paths unless rigorously reviewed, and supports both global deployments and data-zone deployments for residency and sovereignty.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, in your inbox every morning.

Get news like this every morning, free

Context & Analysis

Hugging Face has established itself as the public hub for open-source AI models, with 15 million builders, 400,000 organizations, and over 3 million published models. However, the platform itself has never provided enterprise-grade serving infrastructure — organizations wanting to run these models in production face fragmented choices: license compliance, security review, container image building, runtime selection, CVE patching, and standing up endpoints with observability and governance. This announcement addresses that gap by having Microsoft absorb the operational layer.

The move is significant because it treats open models as full first-class citizens within Foundry's existing architecture. Developers can now mix open-weight and frontier models (from OpenAI, Anthropic, Meta, Mistral, DeepSeek) in a single agent without a separate integration path, using the same Python, C#, JavaScript, and Java SDKs, the same endpoint, and the same billing and observability surface. The weekly refresh cycle and multi-stage curation pipeline — compliance screening, security scanning, runtime optimization, weight validation, and API/performance testing — transfer toil from the developer to Microsoft's platform team, lowering friction for enterprises.

FAQ

What models are included in the Hugging Face Collection on Foundry?: The collection includes trending models from the Hugging Face ecosystem across every modality — text, vision, audio, and multimodal — such as LLMs and VLMs for chat and agents, ASR and speech translation, embeddings, segmentation, and image generation. The catalog is refreshed weekly as the community publishes new models.
How does Microsoft ensure security for these models?: Every model in the collection is security-screened, ships in SafeTensors weight format with no trust_remote_code execution paths unless rigorously reviewed, and Microsoft builds inference container images on supported runtimes that are scanned for CVEs, signed, and published to a Microsoft-managed registry.
What happens to model weights, and where are they stored?: Model weights are pulled from Hugging Face once, validated against the published model card, and stored in Microsoft-managed Azure storage in the regions where the model is served.

Hugging Face BlogRead Original Article

Send on LINE

AI news that matters for your work, delivered every morning.

Pick your industry and the AI tools you use, and get news related to your work every day.

Get it free

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.